reasoning-trace-optimizer

To debug AI agent reasoning errors, you capture interleaved thinking traces between tool calls to expose hidden missteps. Analyzing these reasoning traces detects failure patterns like context degradation and tool confusion, enabling iterative prompt improvements for better reliability.

Interleaved thinking refers to the reasoning steps AI agents take between consecutive tool calls. Failures occur when reasoning traces develop patterns like instruction drift or hallucination, causing the agent to misinterpret tool results and degrade task performance.

Prompt optimization using reasoning trace analysis involves running an iterative loop that detects failure patterns such as context degradation and tool confusion. The workflow outputs an improved prompt, debugging artifacts, and a shareable SKILL.md document.

Common failure patterns in AI agent reasoning traces include context degradation, tool confusion, instruction drift, and hallucination. Detecting these patterns during tool usage allows developers to target specific reasoning errors and generate concrete prompt improvements.

Yes, you can export AI agent debugging logs and optimization artifacts for audit. The optimization workflow saves logs and summaries as artifacts, and generates a shareable SKILL.md capturing the optimization learnings from the reasoning trace analysis.

The best way to fix instruction drift is by analyzing interleaved reasoning traces to identify where the agent deviated from instructions. An iterative optimization loop then generates an improved prompt that corrects the drift and prevents future occurrences.